The AiGovOps Foundation’s front door changed on Thursday. It is a café.
Not a metaphor we chose for a brochure — a decision about what the first minute should feel like. For three months the home page opened on a hero, three doors, four pillars and a toolkit, and a visitor with a real problem had to work out which of our things applied to them. Now there is a table by the door, a practitioner named Wren sitting at it, and one question on the chalkboard:
What brought you in?
Five doors, each a real case
Underneath are five doors, written in the words people actually arrive with. Each is a real case from the hundred verified AI harms we keep in the Library, and each is the start of an hour that ends with a receipt.
- An AI told someone something it shouldn’t have — and now we’re on the hook. Air Canada, 2022. Level 100: a rule in your own words, one decision against it, a real signed receipt made in your browser. Forty-five minutes, nothing installed.
- I have a bot or an agent already, and I’m not sure it’s safe to ship. The $1 Chevy; Knight Capital. Level 200: your rule in CI, refusing the merge on weakening.
- It’s already running, and nobody can say who’s watching it or how to stop it. Arup; Cigna PXDX; nH Predict. Level 300: holds for a named human, and the arithmetic of how long a hold lives before it decays to no.
- Someone needs proof of what the AI actually did — a regulator, a court, a board. Robodebt; Toeslagenaffaire; Horizon. Level 400: verify evidence you did not make, offline, with your name on it.
- I’m not sure what my problem is — I just know AI is in the building. Start with the ten minutes.
That last door is the one most people should take first.
Ten minutes to get started
Run the Gate Check: ten questions about one AI you know, no account, nothing installed. Keep the Gate Card — your verdict and your gaps, never your answers. It is your first receipt. Then pick the door that stung.
The hour that follows is the same at every level. Read the case. Declare the gate — a sentence a
colleague could enforce, or three lines of YAML; both are the same gate and produce the same
receipt. Prove it. Then bring it on Thursday at 15:00, where the first five minutes are for receipts
and a host writes one line: MARK · 100 · <name> · <hash> · host <name> · <date>. That line is the
credential, and the credential is a receipt. Nobody sells it; a stranger can verify it offline.
What we took away
Simplicity was the work, not the decoration. The site is now the café and four rooms — the Library, the open source, the community, the blog — and each room is one path of three steps. Everything else we had built and kept building on top of itself moved to a single resources page. Fewer things, said once.
The engines did not change. Beacon signs, Umbrella declares, Lantern reads, all Apache-2.0, all conforming to OVERT, the open standard from Glacis that this whole practice is built on. What changed is that you no longer have to know any of that to start.
Thursday
This Thursday I am doing my own level 100 — the Air Canada door — and Ken writes the first mark line. The week after, a stranger who has never seen a worksheet tries it unaided; if they finish in under an hour, the release candidate becomes 1.0.0. You will read how both went here, on Tuesday, because that is the one thing we send.
Sit down. What brought you in?